Skip to main content

Token-Sequence Classification

Comparing standard sequence classification approach with a new approach using tokens as labels. The new approach gets better accuracy, but the token approach gets better f1 and roc_auc. The token approach has a shorter runtime.
Created on August 9|Last edited on August 9

  • GoEmotions dataset
  • roberta-base model
  • 2 runs using token approach
  • 2 runs using standard approach


Eval


20304050Step0.520.540.560.58
20304050Step0.40.420.440.46
Run set
4



Training